Hi,

got the problem that houdini isn't properly swapping data to the hard drive when the ram is hitting its limit. Insteat its just crashing. I got a pretty fast M2 SSD and would like to make use of it. How can I make sure this works like it should?

Some operations can not be swapped to disk. Such as the case with FLIP simulations. A single frame of FLIP must fit within physical memory or you will crash, no matter how much free space you have on a drive.

This is why higher amounts of memory allow you to create more detailed effects. 32GB systems are basically the minimum requirement for middle of the road quality effects.

Well, I'm running 64 GB and the crashing occurs during rendertime. Specially volume rasterizing points consumes a lot but I want a certain resolution and not a blobby vdb mess. To render just particles would work but its not looking as good. Also I can use the inbuild vdb motion blur without having to activate it on the mantra rop. Saves some rendertime. So anyway: shouldn't this work with swapping?

Its for such a scene

That's a great looking ship.

Have you cached things out or are you simming while rendering?

Lol

of course I cached out everything. Its not my first sim
I just don't remember houdini crashing a year ago when memory was hitting the limits. Houdini is crashing more since version 16 anyway. When I interrupt a sim I always cross fingers

Just had to ask, sorry.

If you are not running a recent daily build maybe try out one of those. A lot of bugs have been fixed since the production release. But otherwise submit it as a bug.

Maybe a memory chip is going bad..? Or your system is getting too hot? Can you monitor temperature under load?

Rosko Ron
Specially volume rasterizing points consumes a lot but I want a certain resolution and not a blobby vdb mess

Hmmm - it wouldn't be doing this if you have cached properly - just put a file cache node after the ‘volume rasterizing points’

BTW why aren't you using Redshift? IIRC you complained a lot about Mantra rendering in the past.

I'm at 17 minutes per HD frame. Its okay for a CPU render. Rendering all in one go. But thats a good hint, totally didn't think about caching the rasterized points out too.

Houdinis new ocean system isn't very third party renderer friendly I think. So one big disp texture for the whole ocean…well..

Anyway, Redshift is much better for other things. Never would render hard surfaces or motion graphics with mantra…Not everything in 3D consists of heavy VFX shots.


However: thats not explaining why houdini is crashing when it should be swaping insteat. Nothings getting too hot here. Monitoring it all the time.

To diagnose the Mantra issue - you can turn on verbosity on the MantraRop and it will start to spit out memory statistics. You may also want to pack the geometry but not sure if that is needed any more in H16.

Most likely it is swapping already.

Hi, I am having a similar problem when simultaneously simulating and rendering a simple POP network. Is this intended behavior, as with FLIP fluids? If so, where is the documentation stating which parts of Houdini should crash when out of physical memory and which parts of Houdini should start swapping? I am using Mantra, H16 and Ubuntu.

Thanks!

can you please run the top command in a terminal and paste the header info just before it crashes. That will display the vm usage.

ie. on MacOS 16GB physical ram can go to 64GB with swap before stopping, and that may be because of the available space on the HD.

Oh lord, excuse the noise… Ubuntu 17.10 seems to have a problem running both encrypted home directory AND swap files:

https://bugs.launchpad.net/ubuntu/+source/grub2/+bug/1713980 [bugs.launchpad.net]

Which means I had zero swap space available! Sorry for the bogus question!